Data Augmentation Based on Pre-trained Language Model for Event Detection

Event detection (ED) is an important task which needs to identify the event triggers in the sentence and classify the event types. For the general fine-grained event detection task, we propose an event detection scheme based on pre-trained model, combined with data augmentation and pseudo labelling method, which improves the event detection ability of the model. At the same time, we use voting for model ensemble, so as to effectively utilize the advantages of multiple models. Our model achieves F1 score of 69.86% on the test set of CCKS2021 general fine-grained event detection task and ranks the third place in the competition.